Studies were identified using a three-step search strategy. First, an initial limited search was conducted in the MEDLINE database using keywords such as yoga, hypertension, and RCT. The search results were inspected to ensure that the relevant articles were identified. The text words contained in the titles and abstracts of relevant articles and the index terms used to describe the articles were used to develop a search strategy for MEDLINE in consultation with an information specialist/librarian. Second, the search strategy was adapted for other databases in consultation with the information specialist/librarian, and a systematic search across all included databases was undertaken to identify published and unpublished studies. Third, the reference list of previous systematic reviews and studies included in the systematic review were screened for additional studies. The search strategies are reported in the Supplementary File.

The following 16 databases were searched, without date and language restrictions, till March 15, 2021: MEDLINE (from 1946, Ovid), Embase (from 1947, Ovid), CINAHL (from 1937, EBSCO), PsycINFO (from 1806, Ovid), Allied and Complementary Medicine (AMED, from 1985, Ovid), Web of Science (from 1900), Cochrane Central Register of Controlled Trials (CENTRAL, from 1996), Turning Research Into Practice (TRIP, from 1997), AYUSH Research Portal, A Bibliography of Indian Medicine (ABIM), Digital Helpline for Ayurveda Research Articles (DHARA), CAM-QUEST and Directory of Open Access Journals (DOAJ). The search for unpublished studies included OpenGrey (from 1997), EthOS (from 1925), and ProQuest Dissertations and Theses (from 1980, ProQuest).

Following the search, all identified citations were collated and uploaded into Endnote X8.2 (Clarivate Analytics, PA, USA) (39) and duplicates were removed. The remaining records were then imported into Rayyan (Qatar Computing Research Institute [Data Analytics], Doha, Qatar) (40), a web application, to facilitate the title and abstract screening process. Titles and abstracts were independently screened for eligibility using the inclusion criteria by two reviewers. Studies identified as potentially eligible or those without an abstract were retrieved in full. The full-text of these studies was assessed in detail against the inclusion criteria by the two independent reviewers. Full-text studies that did not meet the inclusion criteria were excluded and the reasons for exclusion were reported. Any disagreements that arose between the reviewers were resolved through discussion. If consensus was not reached, then a third reviewer was involved.

Included studies were critically assessed using the standardized JBI critical appraisal tool for RCTs (41). This tool has 13 criteria that can be scored as being met (yes), not met (no), unclear, or not applicable (n/a). One of the criteria was about the blinding of intervention deliverers, which was not possible with yoga interventions and accepted as not applicable. The quality of the individual studies was determined as high quality if 8 or more of criteria scored yes, moderate quality if 5–7 of criteria scored yes, and low quality if 4 or less of criteria scored yes. Two independent reviewers were involved throughout the process and any disagreements arising between reviewers were resolved through discussion. If consensus was not reached, a third reviewer was involved. All studies, regardless of their methodological quality, were included in the review.

Data were extracted using a predeveloped and piloted data extraction form. The data extracted included the characteristics of the studies: author, year of publication, country, final follow-up, participant characteristics (e.g., age, sex, stage of hypertension, medication use, comorbidities, and previous experience of yoga), sample size, intervention and comparator, study outcomes (e.g., blood pressure and adverse events), final follow-up (in weeks), lost to follow-up, and SBP and DBP data extraction time-point (in weeks). In addition, the characteristics of yoga interventions were extracted: intervention development, structure (e.g., session duration, frequency, and intervention duration), delivery characteristics (e.g., the context of the intervention, strategies used to enhance intervention uptake and adherence, and characteristics of yoga instructors), and content (e.g., asana, pranayama, and dhyana and relaxation practices).

The beneficial effects of treatment for most patients with hypertension would be reached within 8 weeks (7, 42). Therefore, the authors extracted the 8-week time point data. Where this time point data was not reported or multiple time points were reported, data from the time point closest to 8 weeks from the start of the intervention were extracted. Intention-to-treat (ITT) data were preferred compared to per-protocol data and post-intervention data preferred over the change from baseline data (i.e., post-intervention score—baseline score). Mean change from baseline was used if only change scores were reported and if there was a significant difference between groups at baseline (≥5 mmHg). However, if studies with a significant difference at baseline did not report change scores, post-intervention data were used, and the imbalance was noted in the risk of bias assessment.

SBP and DBP are continuous data and so mean and SD were extracted. Where no SDs were available, they were calculated from SE, CI, or interquartile ranges, and where no mean scores were available; mean was assumed to be equal to median. In case of missing or unclear data, authors were contacted by e-mail (two times per author).

A tabular and narrative format was used to synthesize results. Some yoga interventions included only one component whereas others incorporated two or three components; therefore, yoga interventions were divided as those incorporating three components of yoga (i.e., yoga interventions incorporating asana, pranayama and dhyana, and relaxation practices), those combining any two components of yoga (e.g., asana and pranayama) and only pranayama interventions. To determine the effectiveness of the studies, a meta-analysis using Review Manager 5.4.1 (43) was conducted as many included RCTs used inappropriate analysis methods to decide the effectiveness of the intervention. First, all yoga interventions were compared with all control groups to show the overall effectiveness of yoga interventions. Then, subgroup analyses were done according to the type of yoga interventions and type of comparators separately for SBP and DBP. Yoga interventions were divided as those incorporating three components of yoga (i.e., yoga interventions incorporating asana, pranayama, and dhyana and relaxation practices), those combining any two components of yoga (e.g., asana and pranayama), and only pranayama interventions. The plan was to divide comparator groups into three as no-intervention, non-pharmaceutical intervention, and pharmaceutical intervention. However, only one study had pharmaceutical intervention as a comparator group. Therefore, comparator groups were divided into two subgroups as no-intervention and non-pharmaceutical intervention. Studies allowing co-medication for both intervention and comparator groups, with no additional intervention, comparator groups were evaluated as no-intervention. All active comparator groups were evaluated as non-pharmaceutical interventions. If yoga interventions were compared with both no intervention and non-pharmaceutical intervention groups within the same study, they were considered separately. Yoga interventions shorter than 8 weeks were separated from the long-term studies in the meta-analysis.

In the meta-analysis, random effects were used due to the heterogeneous nature of the interventions. Statistical heterogeneity was assessed in the meta-analysis using the I² and heterogeneity was considered substantial if I² > 50%. Where there was a sufficient number of studies included in the meta-analysis (at least 10), funnel plots were generated to assess for publication bias.

Thirteen thousand one hundred thirty records were identified through the literature search. After the removal of duplicate records and title and abstract screening, 82 studies were retrieved for full-text screening. Thirty-four RCTs met the inclusion criteria and were included in this systematic review (44–77). (See Figure 1 for PRISMA flowchart) (37) The list of studies found ineligible following full-text review and ongoing RCTs identified from trial registries are presented in the Supplementary File.

See Supplementary Table 1 for the characteristics of the included studies. Thirty-four RCTs met the inclusion criteria. Fifteen studies reported the number of people invited and participated. In total, 5,083 people were invited and 3,125 participated in these studies. Twenty-one RCTs were conducted in India (n = 1,694 participants), 4 in the USA (n = 382) (45, 47, 53, 59), 2 in Germany (n = 415) (46, 67), 2 in Nepal (n = 135) (61, 77) and 1 each in Sweden (n = 191) (49), Thailand (n = 61) (63), Spain (n = 100) (64), Hong-Kong (n = 97) (68) and Brazil (n = 50) (72). Studies recruited participants who were between 18 and 70 years old. Four RCTs recruited only male participants (50, 54, 67, 70). Two recruited only female participants (72, 74) and the sex of the participants was unclear in 3 RCTs (44, 51, 69). Nineteen out of 34 studies did not specify the stages of hypertension. Four studies included only participants with prehypertension, 4 included only participants with stage I hypertension, 5 included participants with prehypertension and stage I hypertension, 1 included participants with stage I and II hypertension and 1 included participants at any stage of hypertension. All the participants were on antihypertensive medication in 12 RCTs, some of the participants were on antihypertensive medication in 5 RCTs and none of the participants was on antihypertensive medication in 7 RCTs. Ten studies did not provide information if participants were on antihypertensive medication. Additional health conditions of the participants included postmenopause (n = 2) (72, 74), type 2 diabetes mellitus (T2DM) (n = 2) (58, 71) and metabolic syndrome (MetS) (n = 1) (68). Previous experience of yoga practice among participants varied across the RCTs and 17 studies recruited yoga-naïve participants whereas 2 studies recruited participants with experience of practicing yoga.

See Supplementary Table 2 for the details of the yoga interventions (e.g., content, structure, and delivery characteristics). Six studies included more than one yoga group and these were considered as separate interventions unless they used the same yogic practice with varying context and delivery characteristics. Consequently, 38 yoga interventions within 34 studies were identified. Of the 38 interventions, 21 incorporated all the three components of yoga (asana, pranayama, and dhyana and relaxation practice); 11 included only pranayama; 3 included pranayama, and dhyana and relaxation practice; 2 included asana and pranayama and one included asana, and dhyana and relaxation practice.

Yoga interventions were compared with a variety of comparators such as no intervention, no additional intervention (where both intervention and control groups received the same intervention e.g., antihypertensive/antidiabetic treatment and lifestyle modification training), lifestyle modification training, exercise, acupuncture, Buteyko breathing, sham breathing, and sham relaxation. Four studies compared yoga intervention with more than one comparator and 38 comparators within 34 studies were identified.

Overall, the studies were assessed as being of low quality (See Supplementary Table 3). The total number of “yes” answers was 7 or more in 6 studies, was 6 in 2 studies, was 5 in 8 studies, and was 4 or less in 18 studies. Even though the design and conduct of studies appeared to be good enough, the methodology was not adequately reported and this resulted in poor scoring. There were poor methodological quality studies as well. Some of the major issues in these studies were: (i) inadequate reporting of the randomization process, of the allocation concealment process, of the blinding of participants and outcome assessors, of whether the study arms were treated identically other than the intervention of interest or not, of the measurement process of outcomes (including adverse events); (ii) imbalance between the treatment groups at baseline; (iii) describing the differences between study arms in terms of their follow up but not analyzing these; (iv) not performing ITT analysis; (v) errors/issues in the sample size calculation and reporting; and (vi) errors/issues in data analysis and reporting (e.g., pre-post analysis and not between the groups).

Adverse events were not reported in 26 RCTs and 6 RCTs reported that no adverse events occurred. 2 RCTs reported adverse events but only one of these studies provided details (46, 59). The study (46) reported that adverse events occurred in yoga and comparator groups but none of these adverse events was judged to be directly related to the yoga intervention.

Funnel plots were generated for yoga intervention vs. control (Supplementary Figures 1, 2) and asana, pranayama, and dhyana and relaxation practice vs. no intervention (Supplementary Figures 3, 4). Funnel plots showed little evidence of publication bias. Meta-analysis including all the yoga interventions showed substantial heterogeneity for SBP (I² = 88%) and DBP (I² = 85%). Subgroup analyses could not reduce heterogeneity substantially. Only two subgroup analyses comparing pranayama with no intervention and non-pharmaceutical interventions for DBP showed no heterogeneity.

Thirty-three out of 34 studies were included in the meta-analysis, comparing yoga with any type of comparators. One of the studies (74) was not included as it compared a yoga intervention with another yoga intervention. The overall effect on SBP and DBP favored yoga interventions compared to comparators (mean difference (MD) −6.49 mmHg; 95% CI −8.94– −4.04; −2.78 mmHg; −4.11– −1.45, respectively) (Figures 2, 3). Of the studies included in the meta-analysis, 16, 13, and 18 interventions were effective in improving SBP, DBP, or either, respectively.

Thirty-eight comparisons were made in subgroup analyses, in total. In subgroup analyses, 18, 14, and 20 out of 38 yoga interventions showed effectiveness in reducing SBP, DBP, or either, respectively. All subgroup comparisons favored yoga interventions except for pranayama vs. non-pharmaceutical intervention in reducing SBP. Asana, pranayama, and dhyana and relaxation practice was more effective in reducing SBP compared to no intervention (−6.71 mmHg; −9.87– −3.55) and non-pharmaceutical interventions (−6.36 mmHg; −11.64– −1.09) (Figures 4, 5). Pranayama was more effective in reducing SBP compared to no intervention (−12.01 mmHg; −20.25– −3.77) but not compared to non-pharmaceutical intervention (−2.39 mmHg; −12.19–7.41) (Figures 6, 7).

All subgroups favored yoga interventions in reducing DBP when compared to no intervention but not when compared to non-pharmaceutical intervention. Asana, pranayama, and dhyana and relaxation practice was more effective in reducing DBP compared to no intervention (−4.67 mmHg; −6.56– −2.77) but not compared to non-pharmaceutical interventions (−2.34 mmHg; −5.42–0.74) (Figures 8, 9). Pranayama was more effective in reducing DBP compared to no intervention (−5.38 mmHg; −8.03– −2.73) but not compared to non-pharmaceutical interventions (1.49 mmHg; −1.28–4.26) (Figures 10, 11). Interventions combining any two components of yoga could not be included in the meta-analysis as no two studies met the criteria to be pooled (46, 49, 58, 64, 67, 72).

Inadequate reporting was found in the included studies in terms of yoga intervention details, study design, and adverse events. Some important aspects of yoga interventions such as the characteristics of yoga providers, the existence of home practice, and attendance and adherence of participants to yoga sessions and home practice were not reported in 32, 59, and 48% of the included studies, respectively. In addition, the intervention development method and the details of the settings where yoga interventions were delivered were not adequately reported. The majority of the studies had low methodological quality with small sample size and short follow-up. The low methodological quality was due to either inadequate reporting or issues in the design of the studies.

As there was no sham or blinded comparison group in the majority of RCTs, the chances of performance bias (of participants) were high. It should be noted that yoga in the west is often considered as a standalone physical activity type of practice, however, traditionally and in many South Asian nations yoga incorporates a healthy lifestyle such as healthy diet, physical activity, smoking cessation, and restriction on alcohol intake (14, 80). Thus, yoga can be considered as a complex intervention, and participants in the yoga study arm might have been motivated to pursue a healthy lifestyle which is conducive to lowering blood pressure.

Future studies should ensure rigorous methodology, i.e., a larger sample size, longer follow-up, and ITT analysis, and reporting should be improved (79). In addition, although blinding of intervention deliverers is not possible in yoga interventions, blinding of participants can be achieved through sham therapies. Adverse events were also not explicitly reported in most of the studies, which are important to indicate the safety of yoga interventions. These issues should be addressed and reporting should be improved in future RCTs.

Despite the low quality and heterogeneity of included studies, our findings suggest yoga interventions may effectively manage hypertension, and provide an indication of the content, structure, and delivery characteristics, which may be most beneficial for managing hypertension. However, most of the included studies have not reported the intervention development process and it is hard to know whether these interventions were carefully thought out (e.g., their safety and acceptability) and comprehensive in their development. In addition, some intervention details were inadequately described (e.g., adherence and home practice), and it is difficult to replicate successful interventions. Thus, future studies should aim to systematically develop a yoga program for hypertension management in consultation with experts (e.g., yoga, hypertension) by using the synthesized findings of the effective interventions in this review. Then, they should aim to evaluate the intervention and implement it, if found effective. In addition, our review, showing that there are some narrative differences between effective and ineffective interventions, paved the way for future research and the next potential step can be to do a meta-analysis, using the methods developed for evaluating effects of individual components in complex interventions (81), to predict what intervention variables are associated with reductions in blood pressure.

To the best of our knowledge, this is the first systematic review to synthesize the content, structure, and delivery characteristics of effective yoga interventions used for managing hypertension and to compare these between effective and ineffective interventions. This systematic review was conducted following a robust systematic process and using the JBI and the PRISMA guidelines by trained JBI accredited systematic reviewers. Everything was done by two reviewers independently with involvement of a third reviewer in case of any disagreements. Sixteen databases were searched with no date or language restrictions. Secondary hypertension and gestational hypertension were outside the scope of this review.

We tended to concentrate on the characteristics of effective yoga interventions and the effectiveness of the interventions was decided based on the statistical significance. This has a potential limitation due to the heterogeneity of the yoga interventions in terms of content, structure, and delivery characteristics and the inclusion of a lot of small studies in this review, which did not have enough power to achieve statistical significance. However, doing a meta-analysis was the best possible way to decide effectiveness because although studies had a comparator group, many of them compared pre-and post-test data within the group, and did not compare between intervention and control groups.